Where systems meet reality.
Architecture, inference, APIs, and the uncomfortable parts of shipping them.
2026
-
AI needs a playground where it can fail
Agents are now good enough to prove their own work end to end. What most teams are missing isn't a better model. It's the foundations: a safe, full-stack place where the agent is allowed to break things.
-
Why more GPUs is not enough for LLM inference
What I learned deploying and tuning large-model inference: KV cache, routing, and cache hierarchy matter as much as raw GPU count.